Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/drivers/block/loop.c | |
3 | * | |
4 | * Written by Theodore Ts'o, 3/29/93 | |
5 | * | |
6 | * Copyright 1993 by Theodore Ts'o. Redistribution of this file is | |
7 | * permitted under the GNU General Public License. | |
8 | * | |
9 | * DES encryption plus some minor changes by Werner Almesberger, 30-MAY-1993 | |
10 | * more DES encryption plus IDEA encryption by Nicholas J. Leon, June 20, 1996 | |
11 | * | |
12 | * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994 | |
13 | * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996 | |
14 | * | |
15 | * Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997 | |
16 | * | |
17 | * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998 | |
18 | * | |
19 | * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998 | |
20 | * | |
21 | * Loadable modules and other fixes by AK, 1998 | |
22 | * | |
23 | * Make real block number available to downstream transfer functions, enables | |
24 | * CBC (and relatives) mode encryption requiring unique IVs per data block. | |
25 | * Reed H. Petty, rhp@draper.net | |
26 | * | |
27 | * Maximum number of loop devices now dynamic via max_loop module parameter. | |
28 | * Russell Kroll <rkroll@exploits.org> 19990701 | |
29 | * | |
30 | * Maximum number of loop devices when compiled-in now selectable by passing | |
31 | * max_loop=<1-255> to the kernel on boot. | |
96de0e25 | 32 | * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999 |
1da177e4 LT |
33 | * |
34 | * Completely rewrite request handling to be make_request_fn style and | |
35 | * non blocking, pushing work to a helper thread. Lots of fixes from | |
36 | * Al Viro too. | |
37 | * Jens Axboe <axboe@suse.de>, Nov 2000 | |
38 | * | |
39 | * Support up to 256 loop devices | |
40 | * Heinz Mauelshagen <mge@sistina.com>, Feb 2002 | |
41 | * | |
42 | * Support for falling back on the write file operation when the address space | |
4e02ed4b | 43 | * operations write_begin is not available on the backing filesystem. |
1da177e4 LT |
44 | * Anton Altaparmakov, 16 Feb 2005 |
45 | * | |
46 | * Still To Fix: | |
47 | * - Advisory locking is ignored here. | |
48 | * - Should use an own CAP_* category instead of CAP_SYS_ADMIN | |
49 | * | |
50 | */ | |
51 | ||
1da177e4 LT |
52 | #include <linux/module.h> |
53 | #include <linux/moduleparam.h> | |
54 | #include <linux/sched.h> | |
55 | #include <linux/fs.h> | |
56 | #include <linux/file.h> | |
57 | #include <linux/stat.h> | |
58 | #include <linux/errno.h> | |
59 | #include <linux/major.h> | |
60 | #include <linux/wait.h> | |
61 | #include <linux/blkdev.h> | |
62 | #include <linux/blkpg.h> | |
63 | #include <linux/init.h> | |
1da177e4 LT |
64 | #include <linux/smp_lock.h> |
65 | #include <linux/swap.h> | |
66 | #include <linux/slab.h> | |
67 | #include <linux/loop.h> | |
863d5b82 | 68 | #include <linux/compat.h> |
1da177e4 | 69 | #include <linux/suspend.h> |
83144186 | 70 | #include <linux/freezer.h> |
1da177e4 LT |
71 | #include <linux/writeback.h> |
72 | #include <linux/buffer_head.h> /* for invalidate_bdev() */ | |
73 | #include <linux/completion.h> | |
74 | #include <linux/highmem.h> | |
75 | #include <linux/gfp.h> | |
6c997918 | 76 | #include <linux/kthread.h> |
d6b29d7c | 77 | #include <linux/splice.h> |
e85a418c | 78 | #include <linux/extent_map.h> |
1da177e4 LT |
79 | |
80 | #include <asm/uaccess.h> | |
81 | ||
73285082 KC |
82 | static LIST_HEAD(loop_devices); |
83 | static DEFINE_MUTEX(loop_devices_mutex); | |
1da177e4 | 84 | |
476a4813 LV |
85 | static int max_part; |
86 | static int part_shift; | |
87 | ||
1da177e4 LT |
88 | /* |
89 | * Transfer functions | |
90 | */ | |
91 | static int transfer_none(struct loop_device *lo, int cmd, | |
92 | struct page *raw_page, unsigned raw_off, | |
93 | struct page *loop_page, unsigned loop_off, | |
94 | int size, sector_t real_block) | |
95 | { | |
96 | char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off; | |
97 | char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off; | |
98 | ||
99 | if (cmd == READ) | |
100 | memcpy(loop_buf, raw_buf, size); | |
101 | else | |
102 | memcpy(raw_buf, loop_buf, size); | |
103 | ||
104 | kunmap_atomic(raw_buf, KM_USER0); | |
105 | kunmap_atomic(loop_buf, KM_USER1); | |
106 | cond_resched(); | |
107 | return 0; | |
108 | } | |
109 | ||
110 | static int transfer_xor(struct loop_device *lo, int cmd, | |
111 | struct page *raw_page, unsigned raw_off, | |
112 | struct page *loop_page, unsigned loop_off, | |
113 | int size, sector_t real_block) | |
114 | { | |
115 | char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off; | |
116 | char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off; | |
117 | char *in, *out, *key; | |
118 | int i, keysize; | |
119 | ||
120 | if (cmd == READ) { | |
121 | in = raw_buf; | |
122 | out = loop_buf; | |
123 | } else { | |
124 | in = loop_buf; | |
125 | out = raw_buf; | |
126 | } | |
127 | ||
128 | key = lo->lo_encrypt_key; | |
129 | keysize = lo->lo_encrypt_key_size; | |
130 | for (i = 0; i < size; i++) | |
131 | *out++ = *in++ ^ key[(i & 511) % keysize]; | |
132 | ||
133 | kunmap_atomic(raw_buf, KM_USER0); | |
134 | kunmap_atomic(loop_buf, KM_USER1); | |
135 | cond_resched(); | |
136 | return 0; | |
137 | } | |
138 | ||
139 | static int xor_init(struct loop_device *lo, const struct loop_info64 *info) | |
140 | { | |
141 | if (unlikely(info->lo_encrypt_key_size <= 0)) | |
142 | return -EINVAL; | |
143 | return 0; | |
144 | } | |
145 | ||
146 | static struct loop_func_table none_funcs = { | |
147 | .number = LO_CRYPT_NONE, | |
148 | .transfer = transfer_none, | |
149 | }; | |
150 | ||
151 | static struct loop_func_table xor_funcs = { | |
152 | .number = LO_CRYPT_XOR, | |
153 | .transfer = transfer_xor, | |
154 | .init = xor_init | |
155 | }; | |
156 | ||
157 | /* xfer_funcs[0] is special - its release function is never called */ | |
158 | static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = { | |
159 | &none_funcs, | |
160 | &xor_funcs | |
161 | }; | |
162 | ||
163 | static loff_t get_loop_size(struct loop_device *lo, struct file *file) | |
164 | { | |
165 | loff_t size, offset, loopsize; | |
166 | ||
167 | /* Compute loopsize in bytes */ | |
168 | size = i_size_read(file->f_mapping->host); | |
169 | offset = lo->lo_offset; | |
170 | loopsize = size - offset; | |
171 | if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize) | |
172 | loopsize = lo->lo_sizelimit; | |
173 | ||
174 | /* | |
175 | * Unfortunately, if we want to do I/O on the device, | |
176 | * the number of 512-byte sectors has to fit into a sector_t. | |
177 | */ | |
178 | return loopsize >> 9; | |
179 | } | |
180 | ||
181 | static int | |
182 | figure_loop_size(struct loop_device *lo) | |
183 | { | |
184 | loff_t size = get_loop_size(lo, lo->lo_backing_file); | |
185 | sector_t x = (sector_t)size; | |
186 | ||
187 | if (unlikely((loff_t)x != size)) | |
188 | return -EFBIG; | |
189 | ||
73285082 | 190 | set_capacity(lo->lo_disk, x); |
1da177e4 LT |
191 | return 0; |
192 | } | |
193 | ||
194 | static inline int | |
195 | lo_do_transfer(struct loop_device *lo, int cmd, | |
196 | struct page *rpage, unsigned roffs, | |
197 | struct page *lpage, unsigned loffs, | |
198 | int size, sector_t rblock) | |
199 | { | |
200 | if (unlikely(!lo->transfer)) | |
201 | return 0; | |
202 | ||
203 | return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); | |
204 | } | |
205 | ||
206 | /** | |
207 | * do_lo_send_aops - helper for writing data to a loop device | |
208 | * | |
209 | * This is the fast version for backing filesystems which implement the address | |
afddba49 | 210 | * space operations write_begin and write_end. |
1da177e4 LT |
211 | */ |
212 | static int do_lo_send_aops(struct loop_device *lo, struct bio_vec *bvec, | |
511de73f | 213 | loff_t pos, struct page *unused) |
1da177e4 LT |
214 | { |
215 | struct file *file = lo->lo_backing_file; /* kudos to NFsckingS */ | |
216 | struct address_space *mapping = file->f_mapping; | |
1da177e4 LT |
217 | pgoff_t index; |
218 | unsigned offset, bv_offs; | |
994fc28c | 219 | int len, ret; |
1da177e4 | 220 | |
1b1dcc1b | 221 | mutex_lock(&mapping->host->i_mutex); |
1da177e4 LT |
222 | index = pos >> PAGE_CACHE_SHIFT; |
223 | offset = pos & ((pgoff_t)PAGE_CACHE_SIZE - 1); | |
224 | bv_offs = bvec->bv_offset; | |
225 | len = bvec->bv_len; | |
226 | while (len > 0) { | |
227 | sector_t IV; | |
afddba49 | 228 | unsigned size, copied; |
1da177e4 | 229 | int transfer_result; |
afddba49 NP |
230 | struct page *page; |
231 | void *fsdata; | |
1da177e4 LT |
232 | |
233 | IV = ((sector_t)index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9); | |
234 | size = PAGE_CACHE_SIZE - offset; | |
235 | if (size > len) | |
236 | size = len; | |
afddba49 NP |
237 | |
238 | ret = pagecache_write_begin(file, mapping, pos, size, 0, | |
239 | &page, &fsdata); | |
240 | if (ret) | |
1da177e4 | 241 | goto fail; |
afddba49 | 242 | |
1da177e4 LT |
243 | transfer_result = lo_do_transfer(lo, WRITE, page, offset, |
244 | bvec->bv_page, bv_offs, size, IV); | |
afddba49 | 245 | copied = size; |
1da177e4 | 246 | if (unlikely(transfer_result)) |
afddba49 NP |
247 | copied = 0; |
248 | ||
249 | ret = pagecache_write_end(file, mapping, pos, size, copied, | |
250 | page, fsdata); | |
8268f5a7 | 251 | if (ret < 0 || ret != copied) |
afddba49 | 252 | goto fail; |
afddba49 NP |
253 | |
254 | if (unlikely(transfer_result)) | |
255 | goto fail; | |
256 | ||
257 | bv_offs += copied; | |
258 | len -= copied; | |
1da177e4 LT |
259 | offset = 0; |
260 | index++; | |
afddba49 | 261 | pos += copied; |
1da177e4 | 262 | } |
994fc28c | 263 | ret = 0; |
1da177e4 | 264 | out: |
1b1dcc1b | 265 | mutex_unlock(&mapping->host->i_mutex); |
1da177e4 | 266 | return ret; |
1da177e4 LT |
267 | fail: |
268 | ret = -1; | |
269 | goto out; | |
270 | } | |
271 | ||
272 | /** | |
273 | * __do_lo_send_write - helper for writing data to a loop device | |
274 | * | |
275 | * This helper just factors out common code between do_lo_send_direct_write() | |
276 | * and do_lo_send_write(). | |
277 | */ | |
858119e1 | 278 | static int __do_lo_send_write(struct file *file, |
98ae6ccd | 279 | u8 *buf, const int len, loff_t pos) |
1da177e4 LT |
280 | { |
281 | ssize_t bw; | |
282 | mm_segment_t old_fs = get_fs(); | |
283 | ||
284 | set_fs(get_ds()); | |
285 | bw = file->f_op->write(file, buf, len, &pos); | |
286 | set_fs(old_fs); | |
287 | if (likely(bw == len)) | |
288 | return 0; | |
289 | printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n", | |
290 | (unsigned long long)pos, len); | |
291 | if (bw >= 0) | |
292 | bw = -EIO; | |
293 | return bw; | |
294 | } | |
295 | ||
296 | /** | |
297 | * do_lo_send_direct_write - helper for writing data to a loop device | |
298 | * | |
299 | * This is the fast, non-transforming version for backing filesystems which do | |
afddba49 | 300 | * not implement the address space operations write_begin and write_end. |
1da177e4 LT |
301 | * It uses the write file operation which should be present on all writeable |
302 | * filesystems. | |
303 | */ | |
304 | static int do_lo_send_direct_write(struct loop_device *lo, | |
511de73f | 305 | struct bio_vec *bvec, loff_t pos, struct page *page) |
1da177e4 LT |
306 | { |
307 | ssize_t bw = __do_lo_send_write(lo->lo_backing_file, | |
98ae6ccd | 308 | kmap(bvec->bv_page) + bvec->bv_offset, |
1da177e4 LT |
309 | bvec->bv_len, pos); |
310 | kunmap(bvec->bv_page); | |
311 | cond_resched(); | |
312 | return bw; | |
313 | } | |
314 | ||
315 | /** | |
316 | * do_lo_send_write - helper for writing data to a loop device | |
317 | * | |
318 | * This is the slow, transforming version for filesystems which do not | |
afddba49 | 319 | * implement the address space operations write_begin and write_end. It |
1da177e4 LT |
320 | * uses the write file operation which should be present on all writeable |
321 | * filesystems. | |
322 | * | |
323 | * Using fops->write is slower than using aops->{prepare,commit}_write in the | |
324 | * transforming case because we need to double buffer the data as we cannot do | |
325 | * the transformations in place as we do not have direct access to the | |
326 | * destination pages of the backing file. | |
327 | */ | |
328 | static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec, | |
511de73f | 329 | loff_t pos, struct page *page) |
1da177e4 LT |
330 | { |
331 | int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page, | |
332 | bvec->bv_offset, bvec->bv_len, pos >> 9); | |
333 | if (likely(!ret)) | |
334 | return __do_lo_send_write(lo->lo_backing_file, | |
98ae6ccd | 335 | page_address(page), bvec->bv_len, |
1da177e4 LT |
336 | pos); |
337 | printk(KERN_ERR "loop: Transfer error at byte offset %llu, " | |
338 | "length %i.\n", (unsigned long long)pos, bvec->bv_len); | |
339 | if (ret > 0) | |
340 | ret = -EIO; | |
341 | return ret; | |
342 | } | |
343 | ||
511de73f | 344 | static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos) |
1da177e4 | 345 | { |
511de73f | 346 | int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t, |
1da177e4 LT |
347 | struct page *page); |
348 | struct bio_vec *bvec; | |
349 | struct page *page = NULL; | |
350 | int i, ret = 0; | |
351 | ||
352 | do_lo_send = do_lo_send_aops; | |
353 | if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) { | |
354 | do_lo_send = do_lo_send_direct_write; | |
355 | if (lo->transfer != transfer_none) { | |
356 | page = alloc_page(GFP_NOIO | __GFP_HIGHMEM); | |
357 | if (unlikely(!page)) | |
358 | goto fail; | |
359 | kmap(page); | |
360 | do_lo_send = do_lo_send_write; | |
361 | } | |
362 | } | |
363 | bio_for_each_segment(bvec, bio, i) { | |
511de73f | 364 | ret = do_lo_send(lo, bvec, pos, page); |
1da177e4 LT |
365 | if (ret < 0) |
366 | break; | |
367 | pos += bvec->bv_len; | |
368 | } | |
369 | if (page) { | |
370 | kunmap(page); | |
371 | __free_page(page); | |
372 | } | |
373 | out: | |
374 | return ret; | |
375 | fail: | |
376 | printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n"); | |
377 | ret = -ENOMEM; | |
378 | goto out; | |
379 | } | |
380 | ||
381 | struct lo_read_data { | |
382 | struct loop_device *lo; | |
383 | struct page *page; | |
384 | unsigned offset; | |
385 | int bsize; | |
386 | }; | |
387 | ||
388 | static int | |
fd582140 JA |
389 | lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
390 | struct splice_desc *sd) | |
1da177e4 | 391 | { |
fd582140 | 392 | struct lo_read_data *p = sd->u.data; |
1da177e4 | 393 | struct loop_device *lo = p->lo; |
fd582140 | 394 | struct page *page = buf->page; |
1da177e4 | 395 | sector_t IV; |
fd582140 JA |
396 | size_t size; |
397 | int ret; | |
1da177e4 | 398 | |
cac36bb0 | 399 | ret = buf->ops->confirm(pipe, buf); |
fd582140 JA |
400 | if (unlikely(ret)) |
401 | return ret; | |
1da177e4 | 402 | |
fd582140 JA |
403 | IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) + |
404 | (buf->offset >> 9); | |
405 | size = sd->len; | |
406 | if (size > p->bsize) | |
407 | size = p->bsize; | |
1da177e4 | 408 | |
fd582140 | 409 | if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) { |
1da177e4 LT |
410 | printk(KERN_ERR "loop: transfer error block %ld\n", |
411 | page->index); | |
fd582140 | 412 | size = -EINVAL; |
1da177e4 LT |
413 | } |
414 | ||
415 | flush_dcache_page(p->page); | |
416 | ||
fd582140 JA |
417 | if (size > 0) |
418 | p->offset += size; | |
419 | ||
1da177e4 LT |
420 | return size; |
421 | } | |
422 | ||
fd582140 JA |
423 | static int |
424 | lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd) | |
425 | { | |
426 | return __splice_from_pipe(pipe, sd, lo_splice_actor); | |
427 | } | |
428 | ||
1da177e4 LT |
429 | static int |
430 | do_lo_receive(struct loop_device *lo, | |
431 | struct bio_vec *bvec, int bsize, loff_t pos) | |
432 | { | |
433 | struct lo_read_data cookie; | |
fd582140 | 434 | struct splice_desc sd; |
1da177e4 | 435 | struct file *file; |
fd582140 | 436 | long retval; |
1da177e4 LT |
437 | |
438 | cookie.lo = lo; | |
439 | cookie.page = bvec->bv_page; | |
440 | cookie.offset = bvec->bv_offset; | |
441 | cookie.bsize = bsize; | |
fd582140 JA |
442 | |
443 | sd.len = 0; | |
444 | sd.total_len = bvec->bv_len; | |
445 | sd.flags = 0; | |
446 | sd.pos = pos; | |
447 | sd.u.data = &cookie; | |
448 | ||
1da177e4 | 449 | file = lo->lo_backing_file; |
fd582140 JA |
450 | retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor); |
451 | ||
452 | if (retval < 0) | |
453 | return retval; | |
454 | ||
455 | return 0; | |
1da177e4 LT |
456 | } |
457 | ||
458 | static int | |
459 | lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos) | |
460 | { | |
461 | struct bio_vec *bvec; | |
462 | int i, ret = 0; | |
463 | ||
464 | bio_for_each_segment(bvec, bio, i) { | |
465 | ret = do_lo_receive(lo, bvec, bsize, pos); | |
466 | if (ret < 0) | |
467 | break; | |
468 | pos += bvec->bv_len; | |
469 | } | |
470 | return ret; | |
471 | } | |
472 | ||
473 | static int do_bio_filebacked(struct loop_device *lo, struct bio *bio) | |
474 | { | |
475 | loff_t pos; | |
476 | int ret; | |
477 | ||
478 | pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset; | |
479 | if (bio_rw(bio) == WRITE) | |
511de73f | 480 | ret = lo_send(lo, bio, pos); |
1da177e4 LT |
481 | else |
482 | ret = lo_receive(lo, bio, lo->lo_blocksize, pos); | |
483 | return ret; | |
484 | } | |
485 | ||
e85a418c JA |
486 | #define __lo_throttle(wq, lock, condition) \ |
487 | do { \ | |
488 | DEFINE_WAIT(__wait); \ | |
489 | for (;;) { \ | |
490 | prepare_to_wait((wq), &__wait, TASK_UNINTERRUPTIBLE); \ | |
491 | if (condition) \ | |
492 | break; \ | |
493 | spin_unlock_irq((lock)); \ | |
494 | wake_up(&lo->lo_event); \ | |
495 | io_schedule(); \ | |
496 | spin_lock_irq((lock)); \ | |
497 | } \ | |
498 | finish_wait((wq), &__wait); \ | |
499 | } while (0) \ | |
500 | ||
501 | #define LO_BIO_THROTTLE 128 | |
502 | #define LO_BIO_THROTTLE_LOW (LO_BIO_THROTTLE / 2) | |
503 | ||
1da177e4 | 504 | /* |
e85a418c JA |
505 | * A normal block device will throttle on request allocation. Do the same |
506 | * for loop to prevent millions of bio's queued internally. | |
507 | */ | |
508 | static void loop_bio_throttle(struct loop_device *lo, struct bio *bio) | |
509 | { | |
510 | __lo_throttle(&lo->lo_bio_wait, &lo->lo_lock, | |
511 | lo->lo_bio_cnt < LO_BIO_THROTTLE); | |
512 | } | |
513 | ||
514 | static void loop_bio_timer(unsigned long data) | |
515 | { | |
516 | struct loop_device *lo = (struct loop_device *) data; | |
517 | ||
518 | wake_up(&lo->lo_event); | |
519 | } | |
520 | ||
521 | /* | |
522 | * Add bio to back of pending list and wakeup thread | |
1da177e4 LT |
523 | */ |
524 | static void loop_add_bio(struct loop_device *lo, struct bio *bio) | |
525 | { | |
e85a418c JA |
526 | loop_bio_throttle(lo, bio); |
527 | ||
1da177e4 LT |
528 | if (lo->lo_biotail) { |
529 | lo->lo_biotail->bi_next = bio; | |
530 | lo->lo_biotail = bio; | |
531 | } else | |
532 | lo->lo_bio = lo->lo_biotail = bio; | |
e85a418c JA |
533 | |
534 | lo->lo_bio_cnt++; | |
535 | ||
536 | smp_mb(); | |
537 | if (lo->lo_bio_cnt > 8) { | |
538 | if (timer_pending(&lo->lo_bio_timer)) | |
539 | del_timer(&lo->lo_bio_timer); | |
540 | ||
541 | if (waitqueue_active(&lo->lo_event)) | |
542 | wake_up(&lo->lo_event); | |
543 | } else if (!timer_pending(&lo->lo_bio_timer)) { | |
544 | lo->lo_bio_timer.expires = jiffies + 1; | |
545 | add_timer(&lo->lo_bio_timer); | |
546 | } | |
1da177e4 LT |
547 | } |
548 | ||
549 | /* | |
550 | * Grab first pending buffer | |
551 | */ | |
552 | static struct bio *loop_get_bio(struct loop_device *lo) | |
553 | { | |
554 | struct bio *bio; | |
555 | ||
1da177e4 LT |
556 | if ((bio = lo->lo_bio)) { |
557 | if (bio == lo->lo_biotail) | |
558 | lo->lo_biotail = NULL; | |
559 | lo->lo_bio = bio->bi_next; | |
560 | bio->bi_next = NULL; | |
561 | } | |
1da177e4 LT |
562 | |
563 | return bio; | |
564 | } | |
565 | ||
e85a418c JA |
566 | static void loop_exit_fastfs(struct loop_device *lo) |
567 | { | |
568 | struct inode *inode = lo->lo_backing_file->f_mapping->host; | |
569 | ||
570 | /* | |
571 | * drop what page cache we instantiated filling holes | |
572 | */ | |
573 | invalidate_inode_pages2(lo->lo_backing_file->f_mapping); | |
574 | ||
575 | blk_queue_ordered(lo->lo_queue, QUEUE_ORDERED_NONE, NULL); | |
576 | ||
577 | mutex_lock(&inode->i_mutex); | |
578 | inode->i_flags &= ~S_SWAPFILE; | |
579 | mutex_unlock(&inode->i_mutex); | |
580 | } | |
581 | ||
582 | static inline u64 lo_bio_offset(struct loop_device *lo, struct bio *bio) | |
583 | { | |
584 | return (u64)lo->lo_offset + ((u64)bio->bi_sector << 9); | |
585 | } | |
586 | ||
587 | /* | |
588 | * Find extent mapping this lo device block to the file block on the real | |
589 | * device | |
590 | */ | |
591 | static struct extent_map *loop_lookup_extent(struct loop_device *lo, | |
592 | u64 offset, gfp_t gfp_mask) | |
593 | { | |
594 | struct address_space *mapping = lo->lo_backing_file->f_mapping; | |
595 | u64 len = 1 << lo->blkbits; | |
596 | ||
597 | return mapping->a_ops->map_extent(mapping, NULL, 0, offset, len, 0, | |
598 | gfp_mask); | |
599 | } | |
600 | ||
601 | static void end_bio_hole_filling(struct bio *bio, int err) | |
602 | { | |
603 | struct address_space *mapping = bio->bi_bdev->bd_inode->i_mapping; | |
604 | struct bio *orig_bio = bio->bi_private; | |
605 | ||
606 | if (mapping->a_ops->extent_io_complete) { | |
607 | u64 start = orig_bio->bi_sector << 9; | |
608 | u64 len = bio->bi_size; | |
609 | ||
610 | mapping->a_ops->extent_io_complete(mapping, start, len); | |
611 | } | |
612 | ||
613 | bio_put(bio); | |
614 | bio_endio(orig_bio, err); | |
615 | } | |
616 | ||
617 | static void fill_extent_hole(struct loop_device *lo, struct bio *bio) | |
618 | { | |
619 | struct address_space *mapping = lo->lo_backing_file->f_mapping; | |
620 | struct bio *new_bio; | |
621 | struct extent_map *em; | |
622 | u64 len = bio->bi_size; | |
623 | u64 start = lo_bio_offset(lo, bio); | |
624 | u64 disk_block; | |
625 | u64 extent_off; | |
626 | ||
627 | /* | |
628 | * change the sector so we can find the correct file offset in our | |
629 | * endio | |
630 | */ | |
631 | bio->bi_sector = start >> 9; | |
632 | ||
633 | mutex_lock(&mapping->host->i_mutex); | |
634 | ||
635 | em = mapping->a_ops->map_extent(mapping, NULL, 0, | |
636 | start, len, 1, GFP_KERNEL); | |
637 | mark_inode_dirty(mapping->host); | |
638 | mutex_unlock(&mapping->host->i_mutex); | |
639 | ||
640 | if (em && !IS_ERR(em)) { | |
641 | disk_block = em->block_start; | |
642 | extent_off = start - em->start; | |
643 | ||
644 | /* | |
645 | * bio_clone() is mempool backed, so if __GFP_WAIT is set | |
646 | * it wont ever fail | |
647 | */ | |
648 | new_bio = bio_clone(bio, GFP_NOIO); | |
649 | new_bio->bi_sector = (disk_block + extent_off) >> 9; | |
650 | new_bio->bi_bdev = em->bdev; | |
651 | new_bio->bi_private = bio; | |
652 | new_bio->bi_size = bio->bi_size; | |
653 | new_bio->bi_end_io = end_bio_hole_filling; | |
654 | free_extent_map(em); | |
655 | ||
656 | generic_make_request(new_bio); | |
657 | } else | |
658 | bio_endio(bio, -EIO); | |
659 | } | |
660 | ||
661 | static void loop_bio_destructor(struct bio *bio) | |
662 | { | |
663 | struct completion *c = (struct completion *) bio->bi_flags; | |
664 | ||
665 | complete(c); | |
666 | } | |
667 | ||
668 | /* | |
669 | * Alloc a hint bio to tell the loop thread to read file blocks for a given | |
670 | * range | |
671 | */ | |
672 | static void loop_schedule_extent_mapping(struct loop_device *lo, | |
673 | struct bio *old_bio) | |
674 | { | |
675 | DECLARE_COMPLETION_ONSTACK(comp); | |
676 | struct bio *bio, stackbio; | |
677 | int do_sync = 0; | |
678 | ||
679 | bio = bio_alloc(GFP_ATOMIC, 0); | |
680 | if (!bio) { | |
681 | bio = &stackbio; | |
682 | bio_init(bio); | |
683 | bio->bi_destructor = loop_bio_destructor; | |
684 | bio->bi_flags = (unsigned long) ∁ | |
685 | do_sync = 1; | |
686 | } | |
687 | ||
688 | bio->bi_rw = LOOP_EXTENT_RW_MAGIC; | |
689 | bio->bi_private = old_bio; | |
690 | ||
691 | loop_add_bio(lo, bio); | |
692 | ||
693 | if (do_sync) { | |
694 | spin_unlock_irq(&lo->lo_lock); | |
695 | wait_for_completion(&comp); | |
696 | spin_lock_irq(&lo->lo_lock); | |
697 | } | |
698 | } | |
699 | ||
700 | static void loop_handle_extent_hole(struct loop_device *lo, struct bio *bio, | |
701 | int sync) | |
702 | { | |
703 | /* | |
704 | * for a read, just zero the data and end the io | |
705 | */ | |
706 | if (bio_data_dir(bio) == READ) { | |
707 | struct bio_vec *bvec; | |
708 | unsigned long flags; | |
709 | int i; | |
710 | ||
711 | bio_for_each_segment(bvec, bio, i) { | |
712 | char *dst = bvec_kmap_irq(bvec, &flags); | |
713 | ||
714 | memset(dst, 0, bvec->bv_len); | |
715 | bvec_kunmap_irq(dst, &flags); | |
716 | } | |
717 | bio_endio(bio, 0); | |
718 | } else { | |
719 | /* | |
720 | * let the page cache handling path do this bio, and then | |
721 | * lookup the mapped blocks after the io has been issued to | |
722 | * instantiate extents. | |
723 | */ | |
724 | if (!sync) | |
725 | loop_add_bio(lo, bio); | |
726 | else | |
727 | fill_extent_hole(lo, bio); | |
728 | } | |
729 | } | |
730 | ||
731 | static inline int lo_is_switch_bio(struct bio *bio) | |
732 | { | |
733 | return !bio->bi_bdev && bio->bi_rw == LOOP_SWITCH_RW_MAGIC; | |
734 | } | |
735 | ||
736 | static inline int lo_is_map_bio(struct bio *bio) | |
737 | { | |
738 | return !bio->bi_bdev && bio->bi_rw == LOOP_EXTENT_RW_MAGIC; | |
739 | } | |
740 | ||
741 | static int __loop_redirect_bio(struct loop_device *lo, struct extent_map *em, | |
742 | struct bio *bio, int sync) | |
743 | { | |
744 | u64 extent_off; | |
745 | u64 disk_block; | |
746 | ||
747 | /* | |
748 | * handle sparse io | |
749 | */ | |
750 | if (em->block_start == EXTENT_MAP_HOLE) { | |
751 | loop_handle_extent_hole(lo, bio, sync); | |
752 | free_extent_map(em); | |
753 | return 0; | |
754 | } | |
755 | ||
756 | /* | |
757 | * not a hole, redirect | |
758 | */ | |
759 | disk_block = em->block_start; | |
760 | extent_off = lo_bio_offset(lo, bio) - em->start; | |
761 | bio->bi_bdev = em->bdev; | |
762 | bio->bi_sector = (disk_block + extent_off) >> 9; | |
763 | free_extent_map(em); | |
764 | return 1; | |
765 | } | |
766 | ||
767 | /* | |
768 | * Change mapping of the bio, so that it points to the real bdev and offset | |
769 | */ | |
770 | static int loop_redirect_bio(struct loop_device *lo, struct bio *bio) | |
771 | { | |
772 | u64 start = lo_bio_offset(lo, bio); | |
773 | struct extent_map *em; | |
774 | ||
775 | em = loop_lookup_extent(lo, start, GFP_ATOMIC); | |
776 | if (IS_ERR(em)) { | |
777 | bio_endio(bio, PTR_ERR(em)); | |
778 | return 0; | |
779 | } else if (!em) { | |
780 | loop_schedule_extent_mapping(lo, bio); | |
781 | return 0; | |
782 | } | |
783 | ||
784 | return __loop_redirect_bio(lo, em, bio, 0); | |
785 | } | |
786 | ||
787 | /* | |
788 | * Wait on bio's on our list to complete before sending a barrier bio | |
789 | * to the below device. Called with lo_lock held. | |
790 | */ | |
791 | static void loop_wait_on_bios(struct loop_device *lo) | |
792 | { | |
793 | __lo_throttle(&lo->lo_bio_wait, &lo->lo_lock, !lo->lo_bio); | |
794 | } | |
795 | ||
796 | static void loop_wait_on_switch(struct loop_device *lo) | |
797 | { | |
798 | __lo_throttle(&lo->lo_bio_wait, &lo->lo_lock, !lo->lo_switch); | |
799 | } | |
800 | ||
165125e1 | 801 | static int loop_make_request(struct request_queue *q, struct bio *old_bio) |
1da177e4 LT |
802 | { |
803 | struct loop_device *lo = q->queuedata; | |
804 | int rw = bio_rw(old_bio); | |
805 | ||
35a82d1a NP |
806 | if (rw == READA) |
807 | rw = READ; | |
808 | ||
809 | BUG_ON(!lo || (rw != READ && rw != WRITE)); | |
1da177e4 LT |
810 | |
811 | spin_lock_irq(&lo->lo_lock); | |
812 | if (lo->lo_state != Lo_bound) | |
35a82d1a NP |
813 | goto out; |
814 | if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY))) | |
815 | goto out; | |
e85a418c JA |
816 | if (lo->lo_flags & LO_FLAGS_FASTFS) { |
817 | /* | |
818 | * If we get a barrier bio, then we just need to wait for | |
819 | * existing bio's to be complete. This can only happen | |
820 | * on the 'new' extent mapped loop, since that is the only | |
821 | * one that supports barriers. | |
822 | */ | |
823 | if (bio_barrier(old_bio)) | |
824 | loop_wait_on_bios(lo); | |
825 | ||
826 | /* | |
827 | * if file switch is in progress, wait for it to complete | |
828 | */ | |
829 | if (!lo_is_switch_bio(old_bio) && lo->lo_switch) | |
830 | loop_wait_on_switch(lo); | |
831 | ||
832 | if (loop_redirect_bio(lo, old_bio)) | |
833 | goto out_redir; | |
834 | goto out_end; | |
835 | } | |
1da177e4 | 836 | loop_add_bio(lo, old_bio); |
35a82d1a | 837 | spin_unlock_irq(&lo->lo_lock); |
1da177e4 | 838 | return 0; |
35a82d1a | 839 | |
1da177e4 | 840 | out: |
6712ecf8 | 841 | bio_io_error(old_bio); |
e85a418c JA |
842 | out_end: |
843 | spin_unlock_irq(&lo->lo_lock); | |
1da177e4 | 844 | return 0; |
e85a418c JA |
845 | |
846 | out_redir: | |
847 | spin_unlock_irq(&lo->lo_lock); | |
848 | return 1; | |
1da177e4 LT |
849 | } |
850 | ||
851 | /* | |
852 | * kick off io on the underlying address space | |
853 | */ | |
165125e1 | 854 | static void loop_unplug(struct request_queue *q) |
1da177e4 LT |
855 | { |
856 | struct loop_device *lo = q->queuedata; | |
857 | ||
75ad23bc | 858 | queue_flag_clear_unlocked(QUEUE_FLAG_PLUGGED, q); |
1da177e4 LT |
859 | blk_run_address_space(lo->lo_backing_file->f_mapping); |
860 | } | |
861 | ||
e85a418c JA |
862 | static void loop_unplug_fastfs(struct request_queue *q) |
863 | { | |
864 | struct loop_device *lo = q->queuedata; | |
865 | struct request_queue *rq = bdev_get_queue(lo->fs_bdev); | |
866 | unsigned long flags; | |
867 | ||
868 | local_irq_save(flags); | |
869 | ||
870 | if (blk_remove_plug(q) && rq->unplug_fn) | |
871 | rq->unplug_fn(rq); | |
872 | ||
873 | local_irq_restore(flags); | |
874 | } | |
875 | ||
1da177e4 LT |
876 | struct switch_request { |
877 | struct file *file; | |
878 | struct completion wait; | |
879 | }; | |
880 | ||
881 | static void do_loop_switch(struct loop_device *, struct switch_request *); | |
e85a418c | 882 | static int loop_init_fastfs(struct loop_device *); |
1da177e4 LT |
883 | |
884 | static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio) | |
885 | { | |
e85a418c JA |
886 | if (lo_is_map_bio(bio)) { |
887 | struct bio *org_bio = bio->bi_private; | |
888 | struct extent_map *em; | |
889 | ||
890 | em = loop_lookup_extent(lo, lo_bio_offset(lo, org_bio), | |
891 | GFP_KERNEL); | |
892 | ||
893 | if (__loop_redirect_bio(lo, em, org_bio, 1)) | |
894 | generic_make_request(org_bio); | |
895 | ||
896 | bio_put(bio); | |
897 | } else if (lo_is_switch_bio(bio)) { | |
1da177e4 LT |
898 | do_loop_switch(lo, bio->bi_private); |
899 | bio_put(bio); | |
900 | } else { | |
e85a418c JA |
901 | if (lo->lo_flags & LO_FLAGS_FASTFS) { |
902 | /* we only get here when filling holes */ | |
903 | fill_extent_hole(lo, bio); | |
904 | } else { | |
905 | int ret = do_bio_filebacked(lo, bio); | |
906 | ||
907 | bio_endio(bio, ret); | |
908 | } | |
1da177e4 LT |
909 | } |
910 | } | |
911 | ||
912 | /* | |
913 | * worker thread that handles reads/writes to file backed loop devices, | |
914 | * to avoid blocking in our make_request_fn. it also does loop decrypting | |
915 | * on reads for block backed loop, as that is too heavy to do from | |
916 | * b_end_io context where irqs may be disabled. | |
6c997918 SH |
917 | * |
918 | * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before | |
919 | * calling kthread_stop(). Therefore once kthread_should_stop() is | |
920 | * true, make_request will not place any more requests. Therefore | |
921 | * once kthread_should_stop() is true and lo_bio is NULL, we are | |
922 | * done with the loop. | |
1da177e4 LT |
923 | */ |
924 | static int loop_thread(void *data) | |
925 | { | |
926 | struct loop_device *lo = data; | |
927 | struct bio *bio; | |
928 | ||
1da177e4 LT |
929 | set_user_nice(current, -20); |
930 | ||
6c997918 | 931 | while (!kthread_should_stop() || lo->lo_bio) { |
09c0dc68 | 932 | |
6c997918 SH |
933 | wait_event_interruptible(lo->lo_event, |
934 | lo->lo_bio || kthread_should_stop()); | |
6c997918 | 935 | if (!lo->lo_bio) |
35a82d1a | 936 | continue; |
35a82d1a | 937 | spin_lock_irq(&lo->lo_lock); |
1da177e4 | 938 | bio = loop_get_bio(lo); |
35a82d1a NP |
939 | spin_unlock_irq(&lo->lo_lock); |
940 | ||
941 | BUG_ON(!bio); | |
e85a418c | 942 | |
1da177e4 | 943 | loop_handle_bio(lo, bio); |
e85a418c JA |
944 | |
945 | spin_lock_irq(&lo->lo_lock); | |
946 | if (--lo->lo_bio_cnt < LO_BIO_THROTTLE_LOW || !lo->lo_bio) | |
947 | wake_up(&lo->lo_bio_wait); | |
948 | spin_unlock_irq(&lo->lo_lock); | |
1da177e4 LT |
949 | } |
950 | ||
1da177e4 LT |
951 | return 0; |
952 | } | |
953 | ||
954 | /* | |
955 | * loop_switch performs the hard work of switching a backing store. | |
956 | * First it needs to flush existing IO, it does this by sending a magic | |
957 | * BIO down the pipe. The completion of this BIO does the actual switch. | |
958 | */ | |
959 | static int loop_switch(struct loop_device *lo, struct file *file) | |
960 | { | |
961 | struct switch_request w; | |
a24eab1e | 962 | struct bio *bio = bio_alloc(GFP_KERNEL, 0); |
1da177e4 LT |
963 | if (!bio) |
964 | return -ENOMEM; | |
965 | init_completion(&w.wait); | |
966 | w.file = file; | |
967 | bio->bi_private = &w; | |
968 | bio->bi_bdev = NULL; | |
e85a418c JA |
969 | bio->bi_rw = LOOP_SWITCH_RW_MAGIC; |
970 | lo->lo_switch = 1; | |
1da177e4 LT |
971 | loop_make_request(lo->lo_queue, bio); |
972 | wait_for_completion(&w.wait); | |
973 | return 0; | |
974 | } | |
975 | ||
14f27939 MB |
976 | /* |
977 | * Helper to flush the IOs in loop, but keeping loop thread running | |
978 | */ | |
979 | static int loop_flush(struct loop_device *lo) | |
980 | { | |
981 | /* loop not yet configured, no running thread, nothing to flush */ | |
982 | if (!lo->lo_thread) | |
983 | return 0; | |
984 | ||
985 | return loop_switch(lo, NULL); | |
986 | } | |
987 | ||
1da177e4 LT |
988 | /* |
989 | * Do the actual switch; called from the BIO completion routine | |
990 | */ | |
991 | static void do_loop_switch(struct loop_device *lo, struct switch_request *p) | |
992 | { | |
993 | struct file *file = p->file; | |
994 | struct file *old_file = lo->lo_backing_file; | |
14f27939 | 995 | struct address_space *mapping; |
e85a418c | 996 | const int fastfs = lo->lo_flags & LO_FLAGS_FASTFS; |
14f27939 MB |
997 | |
998 | /* if no new file, only flush of queued bios requested */ | |
999 | if (!file) | |
1000 | goto out; | |
1da177e4 | 1001 | |
e85a418c JA |
1002 | if (fastfs) |
1003 | loop_exit_fastfs(lo); | |
1004 | ||
14f27939 | 1005 | mapping = file->f_mapping; |
1da177e4 LT |
1006 | mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); |
1007 | lo->lo_backing_file = file; | |
ba52de12 TT |
1008 | lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ? |
1009 | mapping->host->i_bdev->bd_block_size : PAGE_SIZE; | |
1da177e4 LT |
1010 | lo->old_gfp_mask = mapping_gfp_mask(mapping); |
1011 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
e85a418c JA |
1012 | |
1013 | if (fastfs) | |
1014 | loop_init_fastfs(lo); | |
1015 | ||
1016 | lo->lo_switch = 0; | |
1017 | wake_up(&lo->lo_bio_wait); | |
14f27939 | 1018 | out: |
1da177e4 LT |
1019 | complete(&p->wait); |
1020 | } | |
1021 | ||
1022 | ||
1023 | /* | |
1024 | * loop_change_fd switched the backing store of a loopback device to | |
1025 | * a new file. This is useful for operating system installers to free up | |
1026 | * the original file and in High Availability environments to switch to | |
1027 | * an alternative location for the content in case of server meltdown. | |
1028 | * This can only work if the loop device is used read-only, and if the | |
1029 | * new backing store is the same size and type as the old backing store. | |
1030 | */ | |
bb214884 AV |
1031 | static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, |
1032 | unsigned int arg) | |
1da177e4 LT |
1033 | { |
1034 | struct file *file, *old_file; | |
1035 | struct inode *inode; | |
1036 | int error; | |
1037 | ||
1038 | error = -ENXIO; | |
1039 | if (lo->lo_state != Lo_bound) | |
1040 | goto out; | |
1041 | ||
1042 | /* the loop device has to be read-only */ | |
1043 | error = -EINVAL; | |
1044 | if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) | |
1045 | goto out; | |
1046 | ||
1047 | error = -EBADF; | |
1048 | file = fget(arg); | |
1049 | if (!file) | |
1050 | goto out; | |
1051 | ||
1052 | inode = file->f_mapping->host; | |
1053 | old_file = lo->lo_backing_file; | |
1054 | ||
1055 | error = -EINVAL; | |
1056 | ||
1057 | if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) | |
1058 | goto out_putf; | |
1059 | ||
fd582140 JA |
1060 | /* new backing store needs to support loop (eg splice_read) */ |
1061 | if (!inode->i_fop->splice_read) | |
1da177e4 LT |
1062 | goto out_putf; |
1063 | ||
1064 | /* size of the new backing store needs to be the same */ | |
1065 | if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) | |
1066 | goto out_putf; | |
1067 | ||
1068 | /* and ... switch */ | |
1069 | error = loop_switch(lo, file); | |
1070 | if (error) | |
1071 | goto out_putf; | |
1072 | ||
1073 | fput(old_file); | |
476a4813 LV |
1074 | if (max_part > 0) |
1075 | ioctl_by_bdev(bdev, BLKRRPART, 0); | |
1da177e4 LT |
1076 | return 0; |
1077 | ||
1078 | out_putf: | |
1079 | fput(file); | |
1080 | out: | |
1081 | return error; | |
1082 | } | |
1083 | ||
e85a418c JA |
1084 | /* |
1085 | * See if adding this bvec would cause us to spill into a new extent. If so, | |
1086 | * disallow the add to start a new bio. This ensures that the bio we receive | |
1087 | * in loop_make_request() never spans two extents or more. | |
1088 | */ | |
1089 | static int loop_merge_bvec(struct request_queue *q, struct bvec_merge_data *bvm, | |
1090 | struct bio_vec *bvec) | |
1091 | { | |
1092 | struct loop_device *lo = q->queuedata; | |
1093 | struct extent_map *em; | |
1094 | unsigned int ret; | |
1095 | u64 start; | |
1096 | u64 len; | |
1097 | ||
1098 | if (!bvm->bi_size) | |
1099 | return bvec->bv_len; | |
1100 | ||
1101 | start = (u64) lo->lo_offset + ((u64)bvm->bi_sector << 9); | |
1102 | len = bvm->bi_size + bvec->bv_len; | |
1103 | ret = bvec->bv_len; | |
1104 | ||
1105 | em = loop_lookup_extent(lo, start, GFP_ATOMIC); | |
1106 | if (em && !IS_ERR(em)) { | |
1107 | /* | |
1108 | * have extent, disallow if outside that extent | |
1109 | */ | |
1110 | if (start + len > em->start + em->len || start < em->start) | |
1111 | ret = 0; | |
1112 | ||
1113 | free_extent_map(em); | |
1114 | } else | |
1115 | ret = 0; | |
1116 | ||
1117 | return ret; | |
1118 | } | |
1119 | ||
1120 | /* | |
1121 | * Initialize the members pertaining to extent mapping. We will populate | |
1122 | * the tree lazily on demand, as a full scan of a big file can take some | |
1123 | * time. | |
1124 | */ | |
1125 | static int loop_init_fastfs(struct loop_device *lo) | |
1126 | { | |
1127 | struct file *file = lo->lo_backing_file; | |
1128 | struct inode *inode = file->f_mapping->host; | |
1129 | struct request_queue *fs_q; | |
1130 | int ret; | |
1131 | ||
1132 | if (!S_ISREG(inode->i_mode)) | |
1133 | return -EINVAL; | |
1134 | ||
1135 | /* | |
1136 | * Need a working extent_map | |
1137 | */ | |
1138 | if (inode->i_mapping->a_ops->map_extent == NULL) | |
1139 | return -EINVAL; | |
1140 | /* | |
1141 | * invalidate all page cache belonging to this file, it could become | |
1142 | * stale when we directly overwrite blocks. | |
1143 | */ | |
1144 | ret = invalidate_inode_pages2(file->f_mapping); | |
1145 | if (unlikely(ret)) | |
1146 | return ret; | |
1147 | ||
1148 | /* | |
1149 | * disable truncate on this file | |
1150 | */ | |
1151 | mutex_lock(&inode->i_mutex); | |
1152 | inode->i_flags |= S_SWAPFILE; | |
1153 | mutex_unlock(&inode->i_mutex); | |
1154 | ||
1155 | lo->blkbits = inode->i_blkbits; | |
1156 | lo->fs_bdev = file->f_mapping->host->i_sb->s_bdev; | |
1157 | lo->lo_flags |= LO_FLAGS_FASTFS; | |
1158 | lo->lo_queue->unplug_fn = loop_unplug_fastfs; | |
1159 | ||
1160 | blk_queue_merge_bvec(lo->lo_queue, loop_merge_bvec); | |
1161 | blk_queue_ordered(lo->lo_queue, QUEUE_ORDERED_DRAIN, NULL); | |
1162 | ||
1163 | fs_q = bdev_get_queue(lo->fs_bdev); | |
1164 | blk_queue_stack_limits(lo->lo_queue, fs_q); | |
1165 | ||
1166 | printk(KERN_INFO "loop%d: fast redirect\n", lo->lo_number); | |
1167 | return 0; | |
1168 | } | |
1169 | ||
1da177e4 LT |
1170 | static inline int is_loop_device(struct file *file) |
1171 | { | |
1172 | struct inode *i = file->f_mapping->host; | |
1173 | ||
1174 | return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; | |
1175 | } | |
1176 | ||
bb214884 | 1177 | static int loop_set_fd(struct loop_device *lo, fmode_t mode, |
1da177e4 LT |
1178 | struct block_device *bdev, unsigned int arg) |
1179 | { | |
1180 | struct file *file, *f; | |
1181 | struct inode *inode; | |
1182 | struct address_space *mapping; | |
1183 | unsigned lo_blocksize; | |
1184 | int lo_flags = 0; | |
1185 | int error; | |
1186 | loff_t size; | |
1187 | ||
1188 | /* This is safe, since we have a reference from open(). */ | |
1189 | __module_get(THIS_MODULE); | |
1190 | ||
1191 | error = -EBADF; | |
1192 | file = fget(arg); | |
1193 | if (!file) | |
1194 | goto out; | |
1195 | ||
1196 | error = -EBUSY; | |
1197 | if (lo->lo_state != Lo_unbound) | |
1198 | goto out_putf; | |
1199 | ||
1200 | /* Avoid recursion */ | |
1201 | f = file; | |
1202 | while (is_loop_device(f)) { | |
1203 | struct loop_device *l; | |
1204 | ||
bb214884 | 1205 | if (f->f_mapping->host->i_bdev == bdev) |
1da177e4 LT |
1206 | goto out_putf; |
1207 | ||
1208 | l = f->f_mapping->host->i_bdev->bd_disk->private_data; | |
1209 | if (l->lo_state == Lo_unbound) { | |
1210 | error = -EINVAL; | |
1211 | goto out_putf; | |
1212 | } | |
1213 | f = l->lo_backing_file; | |
1214 | } | |
1215 | ||
1216 | mapping = file->f_mapping; | |
1217 | inode = mapping->host; | |
e85a418c | 1218 | lo->lo_flags = 0; |
1da177e4 LT |
1219 | |
1220 | if (!(file->f_mode & FMODE_WRITE)) | |
1221 | lo_flags |= LO_FLAGS_READ_ONLY; | |
1222 | ||
1223 | error = -EINVAL; | |
1224 | if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
f5e54d6e | 1225 | const struct address_space_operations *aops = mapping->a_ops; |
1da177e4 LT |
1226 | /* |
1227 | * If we can't read - sorry. If we only can't write - well, | |
1228 | * it's going to be read-only. | |
1229 | */ | |
fd582140 | 1230 | if (!file->f_op->splice_read) |
1da177e4 | 1231 | goto out_putf; |
4e02ed4b | 1232 | if (aops->write_begin) |
1da177e4 LT |
1233 | lo_flags |= LO_FLAGS_USE_AOPS; |
1234 | if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write) | |
1235 | lo_flags |= LO_FLAGS_READ_ONLY; | |
1236 | ||
ba52de12 TT |
1237 | lo_blocksize = S_ISBLK(inode->i_mode) ? |
1238 | inode->i_bdev->bd_block_size : PAGE_SIZE; | |
1239 | ||
1da177e4 LT |
1240 | error = 0; |
1241 | } else { | |
1242 | goto out_putf; | |
1243 | } | |
1244 | ||
1245 | size = get_loop_size(lo, file); | |
1246 | ||
1247 | if ((loff_t)(sector_t)size != size) { | |
1248 | error = -EFBIG; | |
1249 | goto out_putf; | |
1250 | } | |
1251 | ||
bb214884 | 1252 | if (!(mode & FMODE_WRITE)) |
1da177e4 LT |
1253 | lo_flags |= LO_FLAGS_READ_ONLY; |
1254 | ||
1255 | set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); | |
1256 | ||
1257 | lo->lo_blocksize = lo_blocksize; | |
1258 | lo->lo_device = bdev; | |
1259 | lo->lo_flags = lo_flags; | |
1260 | lo->lo_backing_file = file; | |
eefe85ee | 1261 | lo->transfer = transfer_none; |
1da177e4 LT |
1262 | lo->ioctl = NULL; |
1263 | lo->lo_sizelimit = 0; | |
1264 | lo->old_gfp_mask = mapping_gfp_mask(mapping); | |
1265 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
1266 | ||
1267 | lo->lo_bio = lo->lo_biotail = NULL; | |
1268 | ||
1269 | /* | |
1270 | * set queue make_request_fn, and add limits based on lower level | |
1271 | * device | |
1272 | */ | |
1273 | blk_queue_make_request(lo->lo_queue, loop_make_request); | |
1274 | lo->lo_queue->queuedata = lo; | |
1275 | lo->lo_queue->unplug_fn = loop_unplug; | |
1276 | ||
73285082 | 1277 | set_capacity(lo->lo_disk, size); |
1da177e4 LT |
1278 | bd_set_size(bdev, size << 9); |
1279 | ||
1280 | set_blocksize(bdev, lo_blocksize); | |
1281 | ||
e85a418c JA |
1282 | /* |
1283 | * This needs to be done after setup with another ioctl, | |
1284 | * not automatically like this. | |
1285 | */ | |
1286 | loop_init_fastfs(lo); | |
1287 | ||
6c997918 SH |
1288 | lo->lo_thread = kthread_create(loop_thread, lo, "loop%d", |
1289 | lo->lo_number); | |
1290 | if (IS_ERR(lo->lo_thread)) { | |
1291 | error = PTR_ERR(lo->lo_thread); | |
a7422bf8 | 1292 | goto out_clr; |
6c997918 SH |
1293 | } |
1294 | lo->lo_state = Lo_bound; | |
1295 | wake_up_process(lo->lo_thread); | |
476a4813 LV |
1296 | if (max_part > 0) |
1297 | ioctl_by_bdev(bdev, BLKRRPART, 0); | |
1da177e4 LT |
1298 | return 0; |
1299 | ||
a7422bf8 SH |
1300 | out_clr: |
1301 | lo->lo_thread = NULL; | |
1302 | lo->lo_device = NULL; | |
1303 | lo->lo_backing_file = NULL; | |
1304 | lo->lo_flags = 0; | |
73285082 | 1305 | set_capacity(lo->lo_disk, 0); |
f98393a6 | 1306 | invalidate_bdev(bdev); |
a7422bf8 SH |
1307 | bd_set_size(bdev, 0); |
1308 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask); | |
1309 | lo->lo_state = Lo_unbound; | |
1da177e4 LT |
1310 | out_putf: |
1311 | fput(file); | |
1312 | out: | |
1313 | /* This is safe: open() is still holding a reference. */ | |
1314 | module_put(THIS_MODULE); | |
1315 | return error; | |
1316 | } | |
1317 | ||
1318 | static int | |
1319 | loop_release_xfer(struct loop_device *lo) | |
1320 | { | |
1321 | int err = 0; | |
1322 | struct loop_func_table *xfer = lo->lo_encryption; | |
1323 | ||
1324 | if (xfer) { | |
1325 | if (xfer->release) | |
1326 | err = xfer->release(lo); | |
1327 | lo->transfer = NULL; | |
1328 | lo->lo_encryption = NULL; | |
1329 | module_put(xfer->owner); | |
1330 | } | |
1331 | return err; | |
1332 | } | |
1333 | ||
1334 | static int | |
1335 | loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, | |
1336 | const struct loop_info64 *i) | |
1337 | { | |
1338 | int err = 0; | |
1339 | ||
1340 | if (xfer) { | |
1341 | struct module *owner = xfer->owner; | |
1342 | ||
1343 | if (!try_module_get(owner)) | |
1344 | return -EINVAL; | |
1345 | if (xfer->init) | |
1346 | err = xfer->init(lo, i); | |
1347 | if (err) | |
1348 | module_put(owner); | |
1349 | else | |
1350 | lo->lo_encryption = xfer; | |
1351 | } | |
1352 | return err; | |
1353 | } | |
1354 | ||
1355 | static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev) | |
1356 | { | |
1357 | struct file *filp = lo->lo_backing_file; | |
b4e3ca1a | 1358 | gfp_t gfp = lo->old_gfp_mask; |
1da177e4 LT |
1359 | |
1360 | if (lo->lo_state != Lo_bound) | |
1361 | return -ENXIO; | |
1362 | ||
1363 | if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */ | |
1364 | return -EBUSY; | |
1365 | ||
1366 | if (filp == NULL) | |
1367 | return -EINVAL; | |
1368 | ||
1369 | spin_lock_irq(&lo->lo_lock); | |
1370 | lo->lo_state = Lo_rundown; | |
1da177e4 LT |
1371 | spin_unlock_irq(&lo->lo_lock); |
1372 | ||
6c997918 | 1373 | kthread_stop(lo->lo_thread); |
1da177e4 | 1374 | |
e85a418c JA |
1375 | if (lo->lo_flags & LO_FLAGS_FASTFS) |
1376 | loop_exit_fastfs(lo); | |
1377 | ||
8ae30b89 | 1378 | lo->lo_queue->unplug_fn = NULL; |
1da177e4 LT |
1379 | lo->lo_backing_file = NULL; |
1380 | ||
1381 | loop_release_xfer(lo); | |
1382 | lo->transfer = NULL; | |
1383 | lo->ioctl = NULL; | |
1384 | lo->lo_device = NULL; | |
1385 | lo->lo_encryption = NULL; | |
1386 | lo->lo_offset = 0; | |
1387 | lo->lo_sizelimit = 0; | |
1388 | lo->lo_encrypt_key_size = 0; | |
1389 | lo->lo_flags = 0; | |
6c997918 | 1390 | lo->lo_thread = NULL; |
1da177e4 LT |
1391 | memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); |
1392 | memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); | |
1393 | memset(lo->lo_file_name, 0, LO_NAME_SIZE); | |
bb214884 AV |
1394 | if (bdev) |
1395 | invalidate_bdev(bdev); | |
73285082 | 1396 | set_capacity(lo->lo_disk, 0); |
bb214884 AV |
1397 | if (bdev) |
1398 | bd_set_size(bdev, 0); | |
1da177e4 LT |
1399 | mapping_set_gfp_mask(filp->f_mapping, gfp); |
1400 | lo->lo_state = Lo_unbound; | |
1401 | fput(filp); | |
1402 | /* This is safe: open() is still holding a reference. */ | |
1403 | module_put(THIS_MODULE); | |
476a4813 LV |
1404 | if (max_part > 0) |
1405 | ioctl_by_bdev(bdev, BLKRRPART, 0); | |
1da177e4 LT |
1406 | return 0; |
1407 | } | |
1408 | ||
1409 | static int | |
1410 | loop_set_status(struct loop_device *lo, const struct loop_info64 *info) | |
1411 | { | |
1412 | int err; | |
1413 | struct loop_func_table *xfer; | |
b0fafa81 | 1414 | uid_t uid = current_uid(); |
1da177e4 | 1415 | |
b0fafa81 DH |
1416 | if (lo->lo_encrypt_key_size && |
1417 | lo->lo_key_owner != uid && | |
1da177e4 LT |
1418 | !capable(CAP_SYS_ADMIN)) |
1419 | return -EPERM; | |
1420 | if (lo->lo_state != Lo_bound) | |
1421 | return -ENXIO; | |
1422 | if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) | |
1423 | return -EINVAL; | |
1424 | ||
1425 | err = loop_release_xfer(lo); | |
1426 | if (err) | |
1427 | return err; | |
1428 | ||
1429 | if (info->lo_encrypt_type) { | |
1430 | unsigned int type = info->lo_encrypt_type; | |
1431 | ||
e85a418c JA |
1432 | if (lo->lo_flags & LO_FLAGS_FASTFS) |
1433 | return -EINVAL; | |
1434 | ||
1da177e4 LT |
1435 | if (type >= MAX_LO_CRYPT) |
1436 | return -EINVAL; | |
1437 | xfer = xfer_funcs[type]; | |
1438 | if (xfer == NULL) | |
1439 | return -EINVAL; | |
1440 | } else | |
1441 | xfer = NULL; | |
1442 | ||
e85a418c JA |
1443 | /* |
1444 | * for remaps, offset must be a multiple of full blocks | |
1445 | */ | |
1446 | if ((lo->lo_flags & LO_FLAGS_FASTFS) && | |
1447 | (((1 << lo->blkbits) - 1) & info->lo_offset)) | |
1448 | return -EINVAL; | |
1449 | ||
1da177e4 LT |
1450 | err = loop_init_xfer(lo, xfer, info); |
1451 | if (err) | |
1452 | return err; | |
1453 | ||
1454 | if (lo->lo_offset != info->lo_offset || | |
1455 | lo->lo_sizelimit != info->lo_sizelimit) { | |
1456 | lo->lo_offset = info->lo_offset; | |
1457 | lo->lo_sizelimit = info->lo_sizelimit; | |
1458 | if (figure_loop_size(lo)) | |
1459 | return -EFBIG; | |
1460 | } | |
1461 | ||
1462 | memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); | |
1463 | memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); | |
1464 | lo->lo_file_name[LO_NAME_SIZE-1] = 0; | |
1465 | lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; | |
1466 | ||
1467 | if (!xfer) | |
1468 | xfer = &none_funcs; | |
1469 | lo->transfer = xfer->transfer; | |
1470 | lo->ioctl = xfer->ioctl; | |
1471 | ||
96c58655 DW |
1472 | if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != |
1473 | (info->lo_flags & LO_FLAGS_AUTOCLEAR)) | |
1474 | lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; | |
1475 | ||
1da177e4 LT |
1476 | lo->lo_encrypt_key_size = info->lo_encrypt_key_size; |
1477 | lo->lo_init[0] = info->lo_init[0]; | |
1478 | lo->lo_init[1] = info->lo_init[1]; | |
1479 | if (info->lo_encrypt_key_size) { | |
1480 | memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, | |
1481 | info->lo_encrypt_key_size); | |
b0fafa81 | 1482 | lo->lo_key_owner = uid; |
1da177e4 LT |
1483 | } |
1484 | ||
1485 | return 0; | |
1486 | } | |
1487 | ||
1488 | static int | |
1489 | loop_get_status(struct loop_device *lo, struct loop_info64 *info) | |
1490 | { | |
1491 | struct file *file = lo->lo_backing_file; | |
1492 | struct kstat stat; | |
1493 | int error; | |
1494 | ||
1495 | if (lo->lo_state != Lo_bound) | |
1496 | return -ENXIO; | |
6c648be6 | 1497 | error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat); |
1da177e4 LT |
1498 | if (error) |
1499 | return error; | |
1500 | memset(info, 0, sizeof(*info)); | |
1501 | info->lo_number = lo->lo_number; | |
1502 | info->lo_device = huge_encode_dev(stat.dev); | |
1503 | info->lo_inode = stat.ino; | |
1504 | info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev); | |
1505 | info->lo_offset = lo->lo_offset; | |
1506 | info->lo_sizelimit = lo->lo_sizelimit; | |
1507 | info->lo_flags = lo->lo_flags; | |
1508 | memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); | |
1509 | memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); | |
1510 | info->lo_encrypt_type = | |
1511 | lo->lo_encryption ? lo->lo_encryption->number : 0; | |
1512 | if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { | |
1513 | info->lo_encrypt_key_size = lo->lo_encrypt_key_size; | |
1514 | memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, | |
1515 | lo->lo_encrypt_key_size); | |
1516 | } | |
1517 | return 0; | |
1518 | } | |
1519 | ||
1520 | static void | |
1521 | loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) | |
1522 | { | |
1523 | memset(info64, 0, sizeof(*info64)); | |
1524 | info64->lo_number = info->lo_number; | |
1525 | info64->lo_device = info->lo_device; | |
1526 | info64->lo_inode = info->lo_inode; | |
1527 | info64->lo_rdevice = info->lo_rdevice; | |
1528 | info64->lo_offset = info->lo_offset; | |
1529 | info64->lo_sizelimit = 0; | |
1530 | info64->lo_encrypt_type = info->lo_encrypt_type; | |
1531 | info64->lo_encrypt_key_size = info->lo_encrypt_key_size; | |
1532 | info64->lo_flags = info->lo_flags; | |
1533 | info64->lo_init[0] = info->lo_init[0]; | |
1534 | info64->lo_init[1] = info->lo_init[1]; | |
1535 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1536 | memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); | |
1537 | else | |
1538 | memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); | |
1539 | memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); | |
1540 | } | |
1541 | ||
1542 | static int | |
1543 | loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) | |
1544 | { | |
1545 | memset(info, 0, sizeof(*info)); | |
1546 | info->lo_number = info64->lo_number; | |
1547 | info->lo_device = info64->lo_device; | |
1548 | info->lo_inode = info64->lo_inode; | |
1549 | info->lo_rdevice = info64->lo_rdevice; | |
1550 | info->lo_offset = info64->lo_offset; | |
1551 | info->lo_encrypt_type = info64->lo_encrypt_type; | |
1552 | info->lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1553 | info->lo_flags = info64->lo_flags; | |
1554 | info->lo_init[0] = info64->lo_init[0]; | |
1555 | info->lo_init[1] = info64->lo_init[1]; | |
1556 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1557 | memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1558 | else | |
1559 | memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1560 | memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1561 | ||
1562 | /* error in case values were truncated */ | |
1563 | if (info->lo_device != info64->lo_device || | |
1564 | info->lo_rdevice != info64->lo_rdevice || | |
1565 | info->lo_inode != info64->lo_inode || | |
1566 | info->lo_offset != info64->lo_offset) | |
1567 | return -EOVERFLOW; | |
1568 | ||
1569 | return 0; | |
1570 | } | |
1571 | ||
1572 | static int | |
1573 | loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) | |
1574 | { | |
1575 | struct loop_info info; | |
1576 | struct loop_info64 info64; | |
1577 | ||
1578 | if (copy_from_user(&info, arg, sizeof (struct loop_info))) | |
1579 | return -EFAULT; | |
1580 | loop_info64_from_old(&info, &info64); | |
1581 | return loop_set_status(lo, &info64); | |
1582 | } | |
1583 | ||
1584 | static int | |
1585 | loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) | |
1586 | { | |
1587 | struct loop_info64 info64; | |
1588 | ||
1589 | if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) | |
1590 | return -EFAULT; | |
1591 | return loop_set_status(lo, &info64); | |
1592 | } | |
1593 | ||
1594 | static int | |
1595 | loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { | |
1596 | struct loop_info info; | |
1597 | struct loop_info64 info64; | |
1598 | int err = 0; | |
1599 | ||
1600 | if (!arg) | |
1601 | err = -EINVAL; | |
1602 | if (!err) | |
1603 | err = loop_get_status(lo, &info64); | |
1604 | if (!err) | |
1605 | err = loop_info64_to_old(&info64, &info); | |
1606 | if (!err && copy_to_user(arg, &info, sizeof(info))) | |
1607 | err = -EFAULT; | |
1608 | ||
1609 | return err; | |
1610 | } | |
1611 | ||
1612 | static int | |
1613 | loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { | |
1614 | struct loop_info64 info64; | |
1615 | int err = 0; | |
1616 | ||
1617 | if (!arg) | |
1618 | err = -EINVAL; | |
1619 | if (!err) | |
1620 | err = loop_get_status(lo, &info64); | |
1621 | if (!err && copy_to_user(arg, &info64, sizeof(info64))) | |
1622 | err = -EFAULT; | |
1623 | ||
1624 | return err; | |
1625 | } | |
1626 | ||
bb214884 | 1627 | static int lo_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
1628 | unsigned int cmd, unsigned long arg) |
1629 | { | |
bb214884 | 1630 | struct loop_device *lo = bdev->bd_disk->private_data; |
1da177e4 LT |
1631 | int err; |
1632 | ||
f85221dd | 1633 | mutex_lock(&lo->lo_ctl_mutex); |
1da177e4 LT |
1634 | switch (cmd) { |
1635 | case LOOP_SET_FD: | |
bb214884 | 1636 | err = loop_set_fd(lo, mode, bdev, arg); |
1da177e4 LT |
1637 | break; |
1638 | case LOOP_CHANGE_FD: | |
bb214884 | 1639 | err = loop_change_fd(lo, bdev, arg); |
1da177e4 LT |
1640 | break; |
1641 | case LOOP_CLR_FD: | |
bb214884 | 1642 | err = loop_clr_fd(lo, bdev); |
1da177e4 LT |
1643 | break; |
1644 | case LOOP_SET_STATUS: | |
1645 | err = loop_set_status_old(lo, (struct loop_info __user *) arg); | |
1646 | break; | |
1647 | case LOOP_GET_STATUS: | |
1648 | err = loop_get_status_old(lo, (struct loop_info __user *) arg); | |
1649 | break; | |
1650 | case LOOP_SET_STATUS64: | |
1651 | err = loop_set_status64(lo, (struct loop_info64 __user *) arg); | |
1652 | break; | |
1653 | case LOOP_GET_STATUS64: | |
1654 | err = loop_get_status64(lo, (struct loop_info64 __user *) arg); | |
1655 | break; | |
e85a418c JA |
1656 | case LOOP_SET_FASTFS: |
1657 | err = loop_init_fastfs(lo); | |
1658 | break; | |
1da177e4 LT |
1659 | default: |
1660 | err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; | |
1661 | } | |
f85221dd | 1662 | mutex_unlock(&lo->lo_ctl_mutex); |
1da177e4 LT |
1663 | return err; |
1664 | } | |
1665 | ||
863d5b82 DH |
1666 | #ifdef CONFIG_COMPAT |
1667 | struct compat_loop_info { | |
1668 | compat_int_t lo_number; /* ioctl r/o */ | |
1669 | compat_dev_t lo_device; /* ioctl r/o */ | |
1670 | compat_ulong_t lo_inode; /* ioctl r/o */ | |
1671 | compat_dev_t lo_rdevice; /* ioctl r/o */ | |
1672 | compat_int_t lo_offset; | |
1673 | compat_int_t lo_encrypt_type; | |
1674 | compat_int_t lo_encrypt_key_size; /* ioctl w/o */ | |
1675 | compat_int_t lo_flags; /* ioctl r/o */ | |
1676 | char lo_name[LO_NAME_SIZE]; | |
1677 | unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ | |
1678 | compat_ulong_t lo_init[2]; | |
1679 | char reserved[4]; | |
1680 | }; | |
1681 | ||
1682 | /* | |
1683 | * Transfer 32-bit compatibility structure in userspace to 64-bit loop info | |
1684 | * - noinlined to reduce stack space usage in main part of driver | |
1685 | */ | |
1686 | static noinline int | |
ba674cfc | 1687 | loop_info64_from_compat(const struct compat_loop_info __user *arg, |
863d5b82 DH |
1688 | struct loop_info64 *info64) |
1689 | { | |
1690 | struct compat_loop_info info; | |
1691 | ||
1692 | if (copy_from_user(&info, arg, sizeof(info))) | |
1693 | return -EFAULT; | |
1694 | ||
1695 | memset(info64, 0, sizeof(*info64)); | |
1696 | info64->lo_number = info.lo_number; | |
1697 | info64->lo_device = info.lo_device; | |
1698 | info64->lo_inode = info.lo_inode; | |
1699 | info64->lo_rdevice = info.lo_rdevice; | |
1700 | info64->lo_offset = info.lo_offset; | |
1701 | info64->lo_sizelimit = 0; | |
1702 | info64->lo_encrypt_type = info.lo_encrypt_type; | |
1703 | info64->lo_encrypt_key_size = info.lo_encrypt_key_size; | |
1704 | info64->lo_flags = info.lo_flags; | |
1705 | info64->lo_init[0] = info.lo_init[0]; | |
1706 | info64->lo_init[1] = info.lo_init[1]; | |
1707 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1708 | memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); | |
1709 | else | |
1710 | memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); | |
1711 | memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); | |
1712 | return 0; | |
1713 | } | |
1714 | ||
1715 | /* | |
1716 | * Transfer 64-bit loop info to 32-bit compatibility structure in userspace | |
1717 | * - noinlined to reduce stack space usage in main part of driver | |
1718 | */ | |
1719 | static noinline int | |
1720 | loop_info64_to_compat(const struct loop_info64 *info64, | |
1721 | struct compat_loop_info __user *arg) | |
1722 | { | |
1723 | struct compat_loop_info info; | |
1724 | ||
1725 | memset(&info, 0, sizeof(info)); | |
1726 | info.lo_number = info64->lo_number; | |
1727 | info.lo_device = info64->lo_device; | |
1728 | info.lo_inode = info64->lo_inode; | |
1729 | info.lo_rdevice = info64->lo_rdevice; | |
1730 | info.lo_offset = info64->lo_offset; | |
1731 | info.lo_encrypt_type = info64->lo_encrypt_type; | |
1732 | info.lo_encrypt_key_size = info64->lo_encrypt_key_size; | |
1733 | info.lo_flags = info64->lo_flags; | |
1734 | info.lo_init[0] = info64->lo_init[0]; | |
1735 | info.lo_init[1] = info64->lo_init[1]; | |
1736 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) | |
1737 | memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); | |
1738 | else | |
1739 | memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); | |
1740 | memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); | |
1741 | ||
1742 | /* error in case values were truncated */ | |
1743 | if (info.lo_device != info64->lo_device || | |
1744 | info.lo_rdevice != info64->lo_rdevice || | |
1745 | info.lo_inode != info64->lo_inode || | |
1746 | info.lo_offset != info64->lo_offset || | |
1747 | info.lo_init[0] != info64->lo_init[0] || | |
1748 | info.lo_init[1] != info64->lo_init[1]) | |
1749 | return -EOVERFLOW; | |
1750 | ||
1751 | if (copy_to_user(arg, &info, sizeof(info))) | |
1752 | return -EFAULT; | |
1753 | return 0; | |
1754 | } | |
1755 | ||
1756 | static int | |
1757 | loop_set_status_compat(struct loop_device *lo, | |
1758 | const struct compat_loop_info __user *arg) | |
1759 | { | |
1760 | struct loop_info64 info64; | |
1761 | int ret; | |
1762 | ||
1763 | ret = loop_info64_from_compat(arg, &info64); | |
1764 | if (ret < 0) | |
1765 | return ret; | |
1766 | return loop_set_status(lo, &info64); | |
1767 | } | |
1768 | ||
1769 | static int | |
1770 | loop_get_status_compat(struct loop_device *lo, | |
1771 | struct compat_loop_info __user *arg) | |
1772 | { | |
1773 | struct loop_info64 info64; | |
1774 | int err = 0; | |
1775 | ||
1776 | if (!arg) | |
1777 | err = -EINVAL; | |
1778 | if (!err) | |
1779 | err = loop_get_status(lo, &info64); | |
1780 | if (!err) | |
1781 | err = loop_info64_to_compat(&info64, arg); | |
1782 | return err; | |
1783 | } | |
1784 | ||
bb214884 AV |
1785 | static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, |
1786 | unsigned int cmd, unsigned long arg) | |
863d5b82 | 1787 | { |
bb214884 | 1788 | struct loop_device *lo = bdev->bd_disk->private_data; |
863d5b82 DH |
1789 | int err; |
1790 | ||
863d5b82 DH |
1791 | switch(cmd) { |
1792 | case LOOP_SET_STATUS: | |
1793 | mutex_lock(&lo->lo_ctl_mutex); | |
1794 | err = loop_set_status_compat( | |
1795 | lo, (const struct compat_loop_info __user *) arg); | |
1796 | mutex_unlock(&lo->lo_ctl_mutex); | |
1797 | break; | |
1798 | case LOOP_GET_STATUS: | |
1799 | mutex_lock(&lo->lo_ctl_mutex); | |
1800 | err = loop_get_status_compat( | |
1801 | lo, (struct compat_loop_info __user *) arg); | |
1802 | mutex_unlock(&lo->lo_ctl_mutex); | |
1803 | break; | |
1804 | case LOOP_CLR_FD: | |
1805 | case LOOP_GET_STATUS64: | |
1806 | case LOOP_SET_STATUS64: | |
1807 | arg = (unsigned long) compat_ptr(arg); | |
1808 | case LOOP_SET_FD: | |
1809 | case LOOP_CHANGE_FD: | |
bb214884 | 1810 | err = lo_ioctl(bdev, mode, cmd, arg); |
863d5b82 DH |
1811 | break; |
1812 | default: | |
1813 | err = -ENOIOCTLCMD; | |
1814 | break; | |
1815 | } | |
863d5b82 DH |
1816 | return err; |
1817 | } | |
1818 | #endif | |
1819 | ||
bb214884 | 1820 | static int lo_open(struct block_device *bdev, fmode_t mode) |
1da177e4 | 1821 | { |
bb214884 | 1822 | struct loop_device *lo = bdev->bd_disk->private_data; |
1da177e4 | 1823 | |
f85221dd | 1824 | mutex_lock(&lo->lo_ctl_mutex); |
1da177e4 | 1825 | lo->lo_refcnt++; |
f85221dd | 1826 | mutex_unlock(&lo->lo_ctl_mutex); |
1da177e4 LT |
1827 | |
1828 | return 0; | |
1829 | } | |
1830 | ||
bb214884 | 1831 | static int lo_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 1832 | { |
bb214884 | 1833 | struct loop_device *lo = disk->private_data; |
1da177e4 | 1834 | |
f85221dd | 1835 | mutex_lock(&lo->lo_ctl_mutex); |
96c58655 | 1836 | |
14f27939 MB |
1837 | if (--lo->lo_refcnt) |
1838 | goto out; | |
1839 | ||
1840 | if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { | |
1841 | /* | |
1842 | * In autoclear mode, stop the loop thread | |
1843 | * and remove configuration after last close. | |
1844 | */ | |
bb214884 | 1845 | loop_clr_fd(lo, NULL); |
14f27939 MB |
1846 | } else { |
1847 | /* | |
1848 | * Otherwise keep thread (if running) and config, | |
1849 | * but flush possible ongoing bios in thread. | |
1850 | */ | |
1851 | loop_flush(lo); | |
1852 | } | |
96c58655 | 1853 | |
14f27939 | 1854 | out: |
f85221dd | 1855 | mutex_unlock(&lo->lo_ctl_mutex); |
1da177e4 LT |
1856 | |
1857 | return 0; | |
1858 | } | |
1859 | ||
1860 | static struct block_device_operations lo_fops = { | |
1861 | .owner = THIS_MODULE, | |
bb214884 AV |
1862 | .open = lo_open, |
1863 | .release = lo_release, | |
1864 | .ioctl = lo_ioctl, | |
863d5b82 | 1865 | #ifdef CONFIG_COMPAT |
bb214884 | 1866 | .compat_ioctl = lo_compat_ioctl, |
863d5b82 | 1867 | #endif |
1da177e4 LT |
1868 | }; |
1869 | ||
1870 | /* | |
1871 | * And now the modules code and kernel interface. | |
1872 | */ | |
73285082 | 1873 | static int max_loop; |
1da177e4 | 1874 | module_param(max_loop, int, 0); |
a47653fc | 1875 | MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); |
476a4813 LV |
1876 | module_param(max_part, int, 0); |
1877 | MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); | |
1da177e4 LT |
1878 | MODULE_LICENSE("GPL"); |
1879 | MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); | |
1880 | ||
1881 | int loop_register_transfer(struct loop_func_table *funcs) | |
1882 | { | |
1883 | unsigned int n = funcs->number; | |
1884 | ||
1885 | if (n >= MAX_LO_CRYPT || xfer_funcs[n]) | |
1886 | return -EINVAL; | |
1887 | xfer_funcs[n] = funcs; | |
1888 | return 0; | |
1889 | } | |
1890 | ||
1891 | int loop_unregister_transfer(int number) | |
1892 | { | |
1893 | unsigned int n = number; | |
1894 | struct loop_device *lo; | |
1895 | struct loop_func_table *xfer; | |
1896 | ||
1897 | if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) | |
1898 | return -EINVAL; | |
1899 | ||
1900 | xfer_funcs[n] = NULL; | |
1901 | ||
73285082 | 1902 | list_for_each_entry(lo, &loop_devices, lo_list) { |
f85221dd | 1903 | mutex_lock(&lo->lo_ctl_mutex); |
1da177e4 LT |
1904 | |
1905 | if (lo->lo_encryption == xfer) | |
1906 | loop_release_xfer(lo); | |
1907 | ||
f85221dd | 1908 | mutex_unlock(&lo->lo_ctl_mutex); |
1da177e4 LT |
1909 | } |
1910 | ||
1911 | return 0; | |
1912 | } | |
1913 | ||
1914 | EXPORT_SYMBOL(loop_register_transfer); | |
1915 | EXPORT_SYMBOL(loop_unregister_transfer); | |
1916 | ||
a47653fc | 1917 | static struct loop_device *loop_alloc(int i) |
73285082 KC |
1918 | { |
1919 | struct loop_device *lo; | |
1920 | struct gendisk *disk; | |
1921 | ||
1922 | lo = kzalloc(sizeof(*lo), GFP_KERNEL); | |
1923 | if (!lo) | |
1924 | goto out; | |
1925 | ||
1926 | lo->lo_queue = blk_alloc_queue(GFP_KERNEL); | |
1927 | if (!lo->lo_queue) | |
1928 | goto out_free_dev; | |
1929 | ||
476a4813 | 1930 | disk = lo->lo_disk = alloc_disk(1 << part_shift); |
73285082 KC |
1931 | if (!disk) |
1932 | goto out_free_queue; | |
1933 | ||
1934 | mutex_init(&lo->lo_ctl_mutex); | |
1935 | lo->lo_number = i; | |
1936 | lo->lo_thread = NULL; | |
1937 | init_waitqueue_head(&lo->lo_event); | |
e85a418c JA |
1938 | init_waitqueue_head(&lo->lo_bio_wait); |
1939 | setup_timer(&lo->lo_bio_timer, loop_bio_timer, (unsigned long) lo); | |
73285082 KC |
1940 | spin_lock_init(&lo->lo_lock); |
1941 | disk->major = LOOP_MAJOR; | |
476a4813 | 1942 | disk->first_minor = i << part_shift; |
73285082 KC |
1943 | disk->fops = &lo_fops; |
1944 | disk->private_data = lo; | |
1945 | disk->queue = lo->lo_queue; | |
1946 | sprintf(disk->disk_name, "loop%d", i); | |
73285082 KC |
1947 | return lo; |
1948 | ||
1949 | out_free_queue: | |
1950 | blk_cleanup_queue(lo->lo_queue); | |
1951 | out_free_dev: | |
1952 | kfree(lo); | |
1953 | out: | |
07002e99 | 1954 | return NULL; |
73285082 KC |
1955 | } |
1956 | ||
a47653fc | 1957 | static void loop_free(struct loop_device *lo) |
1da177e4 | 1958 | { |
73285082 KC |
1959 | blk_cleanup_queue(lo->lo_queue); |
1960 | put_disk(lo->lo_disk); | |
1961 | list_del(&lo->lo_list); | |
1962 | kfree(lo); | |
1963 | } | |
1da177e4 | 1964 | |
a47653fc KC |
1965 | static struct loop_device *loop_init_one(int i) |
1966 | { | |
1967 | struct loop_device *lo; | |
1968 | ||
1969 | list_for_each_entry(lo, &loop_devices, lo_list) { | |
1970 | if (lo->lo_number == i) | |
1971 | return lo; | |
1972 | } | |
1973 | ||
1974 | lo = loop_alloc(i); | |
1975 | if (lo) { | |
1976 | add_disk(lo->lo_disk); | |
1977 | list_add_tail(&lo->lo_list, &loop_devices); | |
1978 | } | |
1979 | return lo; | |
1980 | } | |
1981 | ||
1982 | static void loop_del_one(struct loop_device *lo) | |
1983 | { | |
1984 | del_gendisk(lo->lo_disk); | |
1985 | loop_free(lo); | |
1986 | } | |
1987 | ||
73285082 KC |
1988 | static struct kobject *loop_probe(dev_t dev, int *part, void *data) |
1989 | { | |
705962cc | 1990 | struct loop_device *lo; |
07002e99 | 1991 | struct kobject *kobj; |
73285082 | 1992 | |
705962cc AV |
1993 | mutex_lock(&loop_devices_mutex); |
1994 | lo = loop_init_one(dev & MINORMASK); | |
07002e99 | 1995 | kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM); |
73285082 KC |
1996 | mutex_unlock(&loop_devices_mutex); |
1997 | ||
1998 | *part = 0; | |
07002e99 | 1999 | return kobj; |
73285082 KC |
2000 | } |
2001 | ||
2002 | static int __init loop_init(void) | |
2003 | { | |
a47653fc KC |
2004 | int i, nr; |
2005 | unsigned long range; | |
2006 | struct loop_device *lo, *next; | |
2007 | ||
2008 | /* | |
2009 | * loop module now has a feature to instantiate underlying device | |
2010 | * structure on-demand, provided that there is an access dev node. | |
2011 | * However, this will not work well with user space tool that doesn't | |
2012 | * know about such "feature". In order to not break any existing | |
2013 | * tool, we do the following: | |
2014 | * | |
2015 | * (1) if max_loop is specified, create that many upfront, and this | |
2016 | * also becomes a hard limit. | |
2017 | * (2) if max_loop is not specified, create 8 loop device on module | |
2018 | * load, user can further extend loop device by create dev node | |
2019 | * themselves and have kernel automatically instantiate actual | |
2020 | * device on-demand. | |
2021 | */ | |
476a4813 LV |
2022 | |
2023 | part_shift = 0; | |
2024 | if (max_part > 0) | |
2025 | part_shift = fls(max_part); | |
2026 | ||
2027 | if (max_loop > 1UL << (MINORBITS - part_shift)) | |
a47653fc | 2028 | return -EINVAL; |
1da177e4 | 2029 | |
73285082 | 2030 | if (max_loop) { |
a47653fc KC |
2031 | nr = max_loop; |
2032 | range = max_loop; | |
2033 | } else { | |
2034 | nr = 8; | |
476a4813 | 2035 | range = 1UL << (MINORBITS - part_shift); |
a47653fc KC |
2036 | } |
2037 | ||
2038 | if (register_blkdev(LOOP_MAJOR, "loop")) | |
2039 | return -EIO; | |
1da177e4 | 2040 | |
a47653fc KC |
2041 | for (i = 0; i < nr; i++) { |
2042 | lo = loop_alloc(i); | |
2043 | if (!lo) | |
2044 | goto Enomem; | |
2045 | list_add_tail(&lo->lo_list, &loop_devices); | |
1da177e4 | 2046 | } |
a47653fc KC |
2047 | |
2048 | /* point of no return */ | |
2049 | ||
2050 | list_for_each_entry(lo, &loop_devices, lo_list) | |
2051 | add_disk(lo->lo_disk); | |
2052 | ||
2053 | blk_register_region(MKDEV(LOOP_MAJOR, 0), range, | |
2054 | THIS_MODULE, loop_probe, NULL, NULL); | |
2055 | ||
73285082 | 2056 | printk(KERN_INFO "loop: module loaded\n"); |
1da177e4 | 2057 | return 0; |
a47653fc KC |
2058 | |
2059 | Enomem: | |
2060 | printk(KERN_INFO "loop: out of memory\n"); | |
2061 | ||
2062 | list_for_each_entry_safe(lo, next, &loop_devices, lo_list) | |
2063 | loop_free(lo); | |
2064 | ||
2065 | unregister_blkdev(LOOP_MAJOR, "loop"); | |
2066 | return -ENOMEM; | |
1da177e4 LT |
2067 | } |
2068 | ||
73285082 | 2069 | static void __exit loop_exit(void) |
1da177e4 | 2070 | { |
a47653fc | 2071 | unsigned long range; |
73285082 | 2072 | struct loop_device *lo, *next; |
1da177e4 | 2073 | |
476a4813 | 2074 | range = max_loop ? max_loop : 1UL << (MINORBITS - part_shift); |
a47653fc | 2075 | |
73285082 KC |
2076 | list_for_each_entry_safe(lo, next, &loop_devices, lo_list) |
2077 | loop_del_one(lo); | |
2078 | ||
a47653fc | 2079 | blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); |
00d59405 | 2080 | unregister_blkdev(LOOP_MAJOR, "loop"); |
1da177e4 LT |
2081 | } |
2082 | ||
2083 | module_init(loop_init); | |
2084 | module_exit(loop_exit); | |
2085 | ||
2086 | #ifndef MODULE | |
2087 | static int __init max_loop_setup(char *str) | |
2088 | { | |
2089 | max_loop = simple_strtol(str, NULL, 0); | |
2090 | return 1; | |
2091 | } | |
2092 | ||
2093 | __setup("max_loop=", max_loop_setup); | |
2094 | #endif |